Research And Implementation Of Key Techniques Of Content-based Medical Image Retrieval

Along with the popularization of imaging modalities such as CT, MRI, and X-ray, the number of medical images for clinical, education and research is becoming rapid expansion recently. Because of its subjectivity and artificial, texture-based medical image retrieval method could not satisfy the need of large-scale medical image database retrieval. So texture-based medical image retrieval has becoming a hot project in medical image field.Improving the precision and recall of the CBMIR is focused on and the feature extraction and related feedback technology is researched in this thesis. Firstly, introduces the basic theory of CBMIR. Then, proposes a medical image feature extraction method which uses Dual Density Dual-tree complex wavelet. This method can improve the precision and recall. Finally, proposes the related feedback technology based on support vector machine active learning and improve the retrieval performance for the medical images which are the same part of people and different clinicpathological characteristics.The major achievements attained in this thesis are as follows:(1) The key technology of CBMIR is studied and analyzed deeply. The content includes the main low-level features description method for medical image retrieval, the similarity between the image features and the performance evaluation of image retrieval algorithm. It is compared with the classic method through the experiments of the same experiments condition.(2) Due to the complexity of the medical images, proposes a medical image feature extraction method which uses Dual Density Dual-tree complex wavelet. It is compared through multigroup experiments, the precision and recall of this method is higher than the other methods.(3) The current approaches have the limitations when retrievals the images which are the same part of different pathology. Therefore, proposes an active learning technology based on support vector machine for the related feedback. Experiment results demonstrate that after4th related feedback, this method improve the systems’ retrieval average25%inside the top30.(4) A CBMIR system is designed and implemented used VC. Finally, the summary and the prospect on this research are given.